The verdict — the whole site, downloaded and scored
We pulled every URL in anthropic.com's sitemap — 475 pages and posts across 17 sections — downloaded each one, and scored it on the same transparent 20-factor on-page model the rest of the suite uses. The result is uncannily flat: a median on-page score of 59.3, and not one page reaches STRONG. The reason is structural — two defects hit literally every page: nothing on the site declares a date, and almost nothing carries schema. Fix those two at the template level and you lift the entire site at once.
Every page lands in a narrow MODERATE–WEAK band. Uniform mediocrity = a template problem, which is the easiest kind to fix.
Score distribution — all 474 scored pages
none ELITE · none STRONGThe site map — a bird's-eye view
Here is the whole site at a glance. Each tile is a section; its size is the number of pages, its colour is the section's average on-page health. Two sections dominate — News & insights and Research together are 79% of every URL on the site. Click any tile to jump into those pages in the explorer below.
Section proportions
share of all 475 URLsThe site's content investment is overwhelmingly editorial: the News blog and the Research library. That's a strength — but both are exactly the sections with 0% dates and 0% schema, so the biggest content investment gets the least machine-readability. §04 quantifies it.
Crawl coverage — what we actually downloaded
This isn't a sample. We resolved the sitemap, reused the 256 pages we'd already captured in earlier modules (so their scores match exactly), and downloaded the remaining 218 fresh. 474 of 475 URLs scored — the one that didn't is named, not hidden.
Coverage ledger
honest accountingHow it was built
parsed anthropic.com/sitemap.xml → 475 URLs, bucketed into 17 sections.
kept the facts already captured for the core + news pages — identical scores across modules.
fetched every remaining page and parsed its real HTML.
ran the locked 20-factor model on each — structural, keyword-independent.
The two site-wide defects — proven at full scale
The On-Page module flagged these on 31 pages. Now we can prove them across the whole site, and the numbers are stark: they aren't page problems, they're template problems. Both are one-change-fixes-hundreds-of-pages.
No page declares a date
CRITICALdatePublished or dateModifiedFreshness is a query-dependent ranking signal, and AI answer engines preferentially cite dated sources — an undated page reads as undatable. For a site that ships ~10 posts a month, invisibly-fresh content is pure waste. One template change (a visible byline date + Article schema dates) fixes the whole blog and research library.
Almost nothing has schema
CRITICALStructured data is how you assert your entity to Google and the AI crawlers. The product pages that do carry schema are the best-scoring pages on the site — proof the lever works. Inject one Organization block site-wide (sameAs → the Wikidata entity) and Article schema into the editorial templates.
Why the scores are so flat — the proof
best vs worst sectionThe spread from the best section to the worst is only a few points, because the things that vary page-to-page (titles, depth, headings) are already decent — and the things that are uniformly broken (dates, schema, E-E-A-T) are broken everywhere. That's why the fix is a template, not 475 edits.
Section health — every part of the site
The same view, broken out by section so you can see where to aim. Product leads (it's the only section with schema); the giant News and Research libraries sit mid-pack and represent the most pages to gain. Click a row to load that section in the explorer.
| Section | Pages | Avg SEO | Median words | Schema | Dated | No canon. | Best / worst page |
|---|
Freshness — is the site maintained?
From each URL's lastmod in the sitemap, here's the publishing/maintenance signal over time. The site is actively maintained — a steady drumbeat of updates. That's the irony of the date problem: the content is fresh, the pages just never say so where a crawler can read it.
Each bar is the number of URLs the sitemap reports as last-modified that month. A healthy, climbing cadence — the maintenance is real. Surfacing it as on-page dates (§04) converts invisible freshness into a ranking + citation signal.
Authority anchors — where your link equity actually points
Overlaying the Semrush backlink profile onto the map: these are the pages the outside web links to most. The pattern is its own finding — your careers page pulls almost as many links as your homepage, while your defining research sits far down. External authority is concentrated on a handful of pages; the rest of the site has to earn its visibility on-page.
Most-linked pages
referring links · from the backlink profileThe read
A handful of pages hold the external authority. Two moves follow: (1) make sure those high-authority pages link internally to your starved key content (see the Content module), passing equity inward; (2) the deeper backlink gap + outreach queue lives in the Backlinks & Authority module.
Per-page referring counts exist only for the few anchor pages the backlink export surfaces — we show those, and don't fabricate a number for the other 470. Full per-page authority is a backlink-API pull.
Per-page explorer — all 475 URLs
Every page and post on the site, in one place. Search by URL, filter by section, sort any column, and click any row for its full on-page breakdown — the 20-factor score, the three group scores, and the biggest gaps to fix. This is the whole site, auditable.
| URL | Section | SEO | Words | Links | H2 | Schema | Dated |
|---|
Deployable fixes — site-wide leverage
Because the defects are template-level, the fixes are too. Each one below ties a real site-wide number to a single drop-in change that moves hundreds of pages at once — the highest-leverage SEO work available on this site.
Methodology & honest limits
How every number here was produced
The map is structural truth; a few deeper layers need more than a crawl: